[Ohrrpgce] SVN: teeemcee/10051 scons: Attempt to fix mac builds due to --gc-sections being missing
Ralph Versteegen
teeemcee at gmail.com
Tue Feb 20 21:29:11 PST 2018
(Interesting to see the hg revision numbers differ because the build
machines must have other unrelated branches which differed - only the
commit hash is useful)
I tried compiling with clang, but couldn't reproduce this.
However, I notice that the code in optimized_arg() in hspeak for computing
^ is susceptible to rounding error, since subtracting epsilon from the
result of power()
gives a different result. That must be the problem; working on fixing it.
On 21 February 2018 at 16:00, James Paige <Bob at hamsterrepublic.com> wrote:
> The exact version I have on the Mac vm is:
>
> Euphoria to C Translator v4.1.0 development
> 64-bit OS X, Using System Memory
> Revision Date: 2015-02-02 14:18:53, Id: 6300:57179171dbed
>
> So it looks like the same one you have.
>
> On Tuesday, February 20, 2018, Ralph Versteegen <teeemcee at gmail.com>
> wrote:
>
>> Wow! That's pretty alarming.
>>
>> I'm testing with both euphoria 4.1.0 and 4.0.3 on linux, and I get -4097
>> both ways. Maybe this is actually due to using a different C compiler on
>> the Mac: clang. I'll try that next.
>> Alternatively, maybe we're actually using slightly different versions of
>> 4.1.0. On sourceforge there are releases called 4.1.0 and 4.1.0-beta2. The
>> beta2 is actually newer. The exact version I'm using is:
>>
>> Euphoria to C Translator v4.1.0 development
>> 64-bit Linux, Using System Memory
>> Revision Date: 2015-02-02 14:18:53, Id: 5861:57179171dbed
>>
>>
>> ^ is implemented both in hspeak and in FB using floating point and the
>> pow() function, so I'd guess some difference in int <-> float semantics.
>>
>> On 21 February 2018 at 10:16, James Paige <Bob at hamsterrepublic.com>
>> wrote:
>>
>>> So I used hspeak version 3Tg on both
>>> my linux box and my Mac virtual machine, and the resulting autotest.hs
>>> files differed.
>>>
>>> I used the decompile.py tool from your rpgbatch branch of nohrio and
>>> found the difference:
>>>
>>> james at ghostorchid:~/Dropbox/src/otherohr/nohrio/3rdparty$ diff -u
>>> testlinux testmac
>>> diff -u testlinux/32504.hss testmac/32504.hss
>>> --- testlinux/32504.hss 2018-02-20 13:07:44.589047653 -0800
>>> +++ testmac/32504.hss 2018-02-20 13:01:39.493031141 -0800
>>> @@ -1,7 +1,7 @@
>>> Script harmtiletests<-testdamagepatterns()
>>> do(
>>> setcaterpillarmode(1)
>>> - writegeneral(177, readgeneral(177) ,and, -4097)
>>> + writegeneral(177, readgeneral(177) ,and, -4096)
>>> walkhero(0, 3, 1)
>>> waitforhero(0)
>>> _checkpoint()
>>>
>>>
>>> I notice that I am using euphoria 4.0.5 on my Linux box, but 4.1.0 on
>>> the Mac, but none of the warnings at https://rpg.hamsterrepublic.co
>>> m/ohrrpgce/Compiling#hspeak seem to match the problem I see.
>>>
>>>
>>>
>>> On Tue, Feb 20, 2018 at 11:30 AM, James Paige <Bob at hamsterrepublic.com>
>>> wrote:
>>>
>>>> The test cases on Mac started failing with revision 9497, but that is
>>>> just:
>>>>
>>>> r9497 | teeemcee | 2017-09-28 09:00:57 -0700 (Thu, 28 Sep 2017) | 2
>>>> lines
>>>> scons now automatically reimports scripts into autotest.rpg and
>>>> interactivetest.rpg
>>>>
>>>> The failure goes all the way back to revision 8184 which is the
>>>> revision where this particular test was first added:
>>>>
>>>> r8184 | teeemcee | 2016-10-30 07:13:24 -0700 (Sun, 30 Oct 2016) | 4
>>>> lines
>>>> autotest: add (failing) test for harm tiles when there are gaps in the
>>>> active party
>>>> Also add a test.rpg menu option to swap hero slots.
>>>>
>>>> But it only seems to fail if autotest.hss was compiled on mac. If
>>>> compiled by by hspeak on linux, then the problem doesn't happen.
>>>>
>>>> Could this be an hspeak bug?
>>>>
>>>>
>>>> On Tue, Feb 20, 2018 at 8:58 AM, James Paige <Bob at hamsterrepublic.com>
>>>> wrote:
>>>>
>>>>> This does fix compiling on mac
>>>>>
>>>>> I do notice that scons test fails on mac, saying:
>>>>>
>>>>> ==> g_debug.txt <==
>>>>> ! TRACE: rank = 1, slot = 2, get hero stat(slot, stat:hp) = 9, start
>>>>> hp -- harm amount = 10
>>>>> ! TRACE: ASSERT FAILURE:
>>>>> ! TRACE: autotest.hss:2260: get hero stat(slot, stat:hp) == start hp
>>>>> -- harm amount [actual values were 9 == 10]
>>>>> ! TRACE: crash() called from newgame -> runtests1 -> triggertests ->
>>>>> harmtiletests -> testdamagepatterns -> asserthurtpattern -> considerrank ->
>>>>> assertfailure -> crash
>>>>> ! TRACE: in testsuite: assert hurt pattern, test no. 3
>>>>> ! _checkpoint ignored
>>>>> ...Quit the game.
>>>>>
>>>>> I'm going to go back and see if I can figure out when the failure
>>>>> started
>>>>>
>>>>>
>>>>> On Tue, Feb 20, 2018 at 7:54 AM, <subversion at hamsterrepublic.com>
>>>>> wrote:
>>>>>
>>>>>> teeemcee
>>>>>> 2018-02-20 07:54:42 -0800 (Tue, 20 Feb 2018)
>>>>>> 67
>>>>>> scons: Attempt to fix mac builds due to --gc-sections being missing
>>>>>> ---
>>>>>> U wip/SConscript
>>>>>> _______________________________________________
>>>>>> Ohrrpgce mailing list
>>>>>> ohrrpgce at lists.motherhamster.org
>>>>>> http://lists.motherhamster.org/listinfo.cgi/ohrrpgce-motherh
>>>>>> amster.org
>>>>>>
>>>>>
>>>>>
>>>>
>>>
>>> _______________________________________________
>>> Ohrrpgce mailing list
>>> ohrrpgce at lists.motherhamster.org
>>> http://lists.motherhamster.org/listinfo.cgi/ohrrpgce-motherhamster.org
>>>
>>>
>>
> _______________________________________________
> Ohrrpgce mailing list
> ohrrpgce at lists.motherhamster.org
> http://lists.motherhamster.org/listinfo.cgi/ohrrpgce-motherhamster.org
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.motherhamster.org/pipermail/ohrrpgce-motherhamster.org/attachments/20180221/4e1b6822/attachment-0001.html>
More information about the Ohrrpgce
mailing list